Slashdot Mirror
Boston Dynamics' Next-Gen ATLAS Sheds the Tether (roboticstrends.com)
Boston Dynamics' ATLAS robot has been featured here a few times before. An anonymous reader points out that the company has just posted a video of the newest version of the ATLAS, "and it's absolutely incredible."
The video shows ATLAS walk, open a door, maintain its balance while it walks through snow and semi-rough terrain, squat and pick up 10-pound boxes and much more. And it does everything without a tether. The new version is electrically powered and hydraulically actuated. It uses sensors in its body and legs to balance and LIDAR and stereo sensors in its head to avoid obstacles, assess the terrain and help with navigation. This version of Atlas is about 5' 9" tall (about a head shorter than the DRC Atlas) and weighs 180 lbs.
I watched the video with my wife and she was mad at someone being mean to the robot.
Stupid sexy Flanders.
My favorite part was when he deliberately knocked the robot on to it's face. It said good things about it's durability, flexibility, and power density that it was able get back on it's feet. The center of gravity may be behind it to make this easier, which makes you wonder if it can do the same thing if it falls on it's back like a turtle. I would consider "rolling over" to get back up to be a fair tactic there.
It was nice seeing the robot recover from the moving box teasing it. The 2d barcodes made the panic bar door opening less impressive. It looks like the box movement would be improved with better end-effectors for hands, although that is balancing act because many of the high dof end effectors woundn't survive a +200lbs robot landing on them from ~2-3ft drop.
The walk through the snow was very fun to watch. The recoveries from stumbles were pretty solid. I'm looking forward to impovements in energy density and processing speed that allow them to get this thing to run over the same terrain faster than humans. If they can produce a kamikaze bipedal robot for $100,000 that can run over terrain with obstacles and tripping hazards: that would be very useful in an urban combat setting.
Spinning Lidar still represent a significant percentage of that expense, but the servo motors are the real cost driving PITA. Unless you can 3d print or mass produce nice harmonic drive servos for a decent price, this is the primary reason shooting one of these guys full of holes costs $$$. Fortunately, the NVIDIA Tegra X1 has virtually solved the processing side of the equation, although not necessarily within the environmental ratings the DoD wants in its toys.
Yeah, it's funny because I actually felt a bit sorry for the robot. Like it was being picked on.
I also felt this strange sense that it was going to up and whack that guy in the head.
If the robots ever attain sentience and rise, this guy is going to get a beating!
Yes, when you consider the rate of progress over the past decade and project it out a decade into the future, it feels like a revolution is happening. Humanoid robots have been promised for a long time, but they turned out to be much more difficult that initially thought. Turns out that 'much more difficult' doesn't mean impossible and continued advances in sensors, actuators, materials, computational hardware, and algorithms are finally breaking through. Engineering advances will continue and near human capabilities will become super-human capabilities in some areas. Of course humanoid is only the most relatable versions as other kinds of robots have been transforming our lives for decades.
I'm sure the P/W ratio for today's bipedal autonomous robots is lower than that of humans. But the Wikipedia article on the ratio cites 20W/kg for human cyclists as a 5-second maximum, and 174W/kg for a Tesla Roadster -- and that's the whole freaking car, motors, batteries, chassis, body, upholstery, the works.
Batteries still are, and will probably remain, a strong limiting factor. But not motors -- for example, Siemens announced a 50kg electric aircraft motor that delivers 260kw continuous output power. That's five KILOWATTS per kg. Without active cooling, that motor would melt quickly at full power, but scale it down to a 5kg "leg muscle" delivering 26kw impulse power, and yeah, you're heavily outperforming a human leg muscle.
At least, I think so. Disclaimer: I'm no biomechanical engineer.